Payment System: What Actually Moves the Money Underneath

A payment system is the shared arrangement underneath every payment application and every card: the banks hold accounts at a common place, what they owe each other is worked out, and money moves between those accounts. Setu Payments Limited handled 1,200 crore instructions in one stated year, and Rs 3,60,000 crore rode on them, so the average was Rs 300.00/-. The average of Rs 300.00/- decides how the system gets built.

The four parties that stand around a payment have all been parties that handle an instruction. Somebody takes the instruction off a phone. Somebody carries it. Somebody signs the business up so that it can receive one at all. Somebody wrote the rule book that lets a card issued by one bank work at a business banking with another. All of that is real work and none of it is the subject here.

Underneath all of them is one arrangement that almost nobody looks at, and it is the only place in the entire chain where money changes hands. Almost everything a payment system spends is driven by how many instructions it handles, and almost nothing by how much value is written on them. A message about a small payment costs what a message about a large one costs. A record of a small payment costs what a record of a large one costs. A query somebody asks six weeks later costs the same either way. So the value carried shows the purpose of the system, and the instruction count shows what it costs and what it must withstand on its worst evening.

What is a payment system, once everything with a screen is stripped away?

The test is short enough to carry around and it settles the question every time. If a thing has a screen, it is not the payment system. The screen is something standing on the payment system.

An application on a phone is a way of entering an instruction. A payment gatewayThe party whose job is to carry an instruction from wherever it was entered to wherever it will be acted on. Which parties are allowed to do this, and what each may touch on the way, is worked separately. is a way of carrying that instruction from where it was entered to where it will be acted on. A merchant acquirerWhoever takes a shop or a service onto its books so that payments can reach it at all, and then stands behind those sales to everyone else in the chain. Worked separately. is the party that signs a business up in the first place. A payment aggregatorA party that takes on many businesses' payments as a bundle and settles with each of them, rather than every business dealing with a bank on its own. Worked separately. takes many businesses on as a bundle. A card networkThe body whose rule book lets a card issued by one bank be accepted by a business banking with another. Worked separately. writes the rule book those parties work inside. Every single one of them passes instructions about money and not one of them moves any.

Underneath the whole stack sits a shared arrangement that a set of banks are members of, where the obligations between those banks are worked out and then discharged, and it is the only layer at which a balance changes. Think of a large residential block with a caretaker's office in the basement. Every flat has its own front door, its own bell and its own way of taking a delivery, and all of that is what a resident sees. But the block has one water meter, one electricity connection and one account through which every flat's share is squared up at the end of the month. The doorbells are where the instructions arrive. The basement is where the money moves.

How Payment Systems Move Money: where in the chain does a balance change?

Here is the mechanism in order, and it is only four steps long. The third one is the answer to a question this whole sequence keeps asking, so read them slowly.

First, every member bank holds an account at a common place. The account is the bank's own, it carries a balance, and it exists for exactly one purpose. The account is debited and credited when obligations between members are met. Second, instructions arrive saying that a customer of one member bank is paying a customer of another. Thousands of them, or crores of them, arriving from every application and gateway sitting above. Third, what each bank owes each other bank is worked out. Fourth, at whatever moment the arrangement provides for, the accounts at the common place are debited and credited so that those obligations are discharged.

A balance in an account at a common place moves, and that debit and credit is the only moment in the entire chain when money actually moves. Everything before it is a message saying an obligation exists. Everything after it is a record of one that has been met. The payer never sees the moment, the payee never sees it, and no screen anywhere reports it. Nobody watches the money move, so plenty of people believe a confirmation is the money.

Take the household version. Two neighbours run a shared kitchen through the month, and one keeps a notebook. Every time one buys something for the other, a line goes in the book. The lines are instructions and the notebook is full of them by the twenty-eighth. Nothing whatever has moved. Then somebody adds up both columns, works out who is behind, and hands over a single note. The handover of that single note is the settlement. Until it happens, every line in the book is an obligation that exists and has not been met.

Why does almost nothing leave the system when one person pays another?

One fact makes everything after it possible, and it takes one sentence to state and a moment to absorb. When a customer of one bank pays a customer of another bank, nothing leaves the banking system. One claim shrinks and another grows. The rupees do not go anywhere; they are relabelled.

Follow the consequence. If money is not leaving, then between any two banks the payments run in both directions all day. Bank one's customers are paying bank two's customers, and bank two's customers are paying bank one's customers, and most of what each side owes is met by what the other side owes. Because the money stays inside, only the difference between what two banks owe each other has to be settled, and that difference is very much smaller than the value of the instructions that produced it.

The everyday version is a mall with ten shops in it. The tailor buys lunch from the food stall. The stall buys a shirt from the tailor. The chemist buys tea from the stall and the stall buys medicine from the chemist. If every one of those was squared up in cash the moment it happened, somebody would be walking around the mall all day with a bag of notes. Sensible shopkeepers keep a running tally instead, and move only what is left over at the end.

The drawing below puts six made up instructions among three banks and shows the shrink. Six instructions running to Rs 3,750/- leave Rs 250/- to move. The settled figure is fifteen times smaller. The multiple of fifteen belongs to those six instructions and to nothing else in the world. Change any one of them and it changes. The direction never changes. The settled figure is always the smaller one, and it is smaller because payments run both ways.

What are the two shapes a settlement can take, and what does each buy?

There are two shapes, and each one is known by its shape rather than by anybody's name. Which named arrangement has which shape is a separate question, and every hour and interval attached to it is set by the authority.

The first shape gathers instructions over an interval, works out what each side owes at the end of it, and makes one movement that discharges many instructions at once. The second shape discharges every instruction on its own, in full, as it arrives. The interval is the whole of the difference, and everything else about the two arrangements follows from it.

Gathering and netting buys cheapness at the price of an interval in which obligations sit undischarged, and settling one at a time removes that interval at the price of doing the expensive thing on every single instruction. It is one trade read in two directions, which is why a reader who understands it in one direction usually gets the other one wrong. The expensive thing is the working out and the movement. Done once for a whole gathered round, it is spread across everything in that round. Done once for each instruction, it is paid in full each time.

The interval holds something worth naming plainly, and that something is the reason the second shape exists at all. Between the moment an instruction is accepted and the moment the balances move, an obligation exists between two banks that has not been met. If something goes badly wrong inside that window, that obligation is what is exposed. Its size is the size of what is sitting in the window. The undischarged obligation is the hinge of everything that follows. How long any such window is, when a cut-offThe moment after which an arriving instruction falls into the next round rather than the current one. Every such moment is set by the authority. falls and how often an arrangement settles are all set by the authority and none of them appears here.

How Payment-Rail Economics Work: what does a system spend on, and what drives each line?

Three drivers, and there are only three. Each one is named with what it follows, and no amount is attached to any of them. The authority sets what a payment system spends and what it may charge the banks that are its members.

Per instruction, per settlement and the capacity built for the busiest moment are the three things a payment system's spending hangs from, and the value carried drives none of them. Take them one at a time.

Per instruction covers the message that carries it, the checks run on it, the record written of it and the query somebody asks about it seven weeks later when a payment they thought had gone through has not appeared. Every one of those is done once for each instruction and its cost follows the count. Per settlement covers the working out of what each bank owes each other bank and the movement that discharges it. Per settlement follows how often the arrangement settles, not how many instructions were gathered into the round: a round with a hundred instructions in it and a round with a hundred crore in it are one settlement each. Capacity covers the machines, the lines and the people, and it is built for the busiest moment the system has to keep working through.

Now look at the list again and notice what is missing from it. The value carried appears in none of the three. Not in the messages, not in the records, not in the settlements, not in the peak. A message saying Rs 60/- is going from one account to another is the same message as one saying Rs 60,000/- is. The message is the same length, it takes the same checks, it produces the same record and it occupies the same fraction of a heavy hour.

Why does the average instruction decide everything else?

Setu Payments Limited, an invented payment system, hands over exactly two numbers about one year and holds nothing else whatever. The system put through 1,200 crore instructions across that year, and the money on those instructions came to Rs 3,60,000 crore. Take the larger of those and divide it by the smaller. Rs 3,60,000 crore over 1,200 crore instructions is Rs 300.00/- an instruction. Multiply Rs 300.00/- back by 1,200 crore instructions and Rs 3,60,000 crore comes back, the only check the figure needs.

The value carried shows the purpose of the system, and the instruction count shows what it costs and what it has to withstand, and neither of those two readings substitutes for the other. A reader who has only the value has a number that sounds like size and drives nothing. A reader who has only the count knows what the system spends on and has no idea what it is for. Both are needed, and both together also give the average. The average is the single most informative number about a payment system that anybody publishes.

Here is why it matters so much. A system whose average payment is three hundred rupees is doing something entirely different from one whose average is three lakh, and the economics of each follow from that one number. Three hundred rupees is a chemist, an autorickshaw, a plate of food, a school fee instalment. Three lakh is a bank paying another bank, a business settling with a supplier, somebody buying a car. A chemist's bill and a bank's payment to another bank are not the same activity, they are not carried by the same shape of arrangement, and the reason they are not is arithmetic rather than preference.

One thing the average is not, and this needs saying because it is where readers are most quietly misled. Rs 300.00/- is a quotient. Dividing the value by the count produces a fact about how the whole arrangement was used, and never a picture of what any one person paid. One count and one value are all Setu Payments Limited reports. There is no middle instruction, no spread, no largest and no smallest. Nobody may say from it that a typical payment is Rs 300.00/-, and nobody may say most instructions are small. The same average comes out of a crowd of small instructions, and equally out of a very few huge ones surrounded by an ocean of tiny ones, and those two systems have almost nothing in common.

Why does a small average payment force netting rather than the other shape?

Work it as a consequence rather than taking it on trust. Settling an instruction on its own is the expensive shape, and the reason it is expensive is that the working out and the movement have to happen once for that instruction alone. Crucially, the work is the same work, so that cost is about the same whether the instruction carries Rs 60/- or Rs 6,00,000/-.

Now run that shape on Setu Payments Limited's year. Twelve hundred crore instructions, each carrying Rs 300.00/- on average. Doing the expensive thing once for each of 1,200 crore instructions means paying for 1,200 crore settlements against a very small amount of value each time. A system with a small average payment has to find a way to pay for one settlement across a great many instructions, and finding that way is exactly what netting is.

The stall outside a college gate is the same problem. If the person running it had to walk to the bank and back for every plate of food sold, the walk would cost more than the plate. The stallholder takes the day's sales together instead and makes one trip. Nobody chose that arrangement because it was elegant. The cost of the trip does not care how much is in the bag, and the amount in the bag from any single sale would never cover it.

Why does a large average payment force the opposite shape?

The opposite direction is the one readers do not anticipate, and both halves of the trade turn over at once.

Start with the cost. On an instruction carrying Rs 3,00,000/-, the cost of settling it on its own is trivial beside the amount involved. A settlement that was unaffordable at Rs 300.00/- an instruction is barely noticeable at Rs 3,00,000/-. The cost did not change and the amount it is set against went up a thousandfold. The same Rs 3,60,000 crore placed onto instructions averaging Rs 3,00,000/- each needs only 1.2 crore of them, a thousandth of the reported 1,200 crore. The average is larger by that same factor of a thousand for exactly that reason.

Now the other half. The interval that netting creates, in which an obligation exists and has not been discharged, is an exposure, and the size of that exposure is the size of what is sitting in the window. At Rs 300.00/- an instruction that window holds small things. At Rs 3,00,000/- an instruction it holds large ones, and it holds them between banks rather than between people.

A settlement that is unaffordable at a small average is cheap at a large one, and an interval that is tolerable at a small average stops being tolerable at a large one, so the two systems end up with opposite architectures for the same reason read in two directions. Neither shape is better than the other. The two shapes are answers to different questions, and the question is set by the average.

What has to be built for the busiest moment, and is that a value or a count?

A payment system has to keep working on its heaviest hour, and it has to be built for that hour rather than for the average one. So the question worth being exact about is what makes an hour heavy.

Capacity is built for a count, and a system could carry ten times the value on the same instructions with almost nothing added. What makes an hour heavy is instructions arriving. More people, all at once, pressing pay. Not more rupees on the instructions of the same people.

Think about the stall outside one office building at half past one. The stall is not busier because somebody bought something expensive. More people came, and what runs out is the speed of handing plates over, a count rather than an amount. If one customer that afternoon spent ten times what anybody else did, the queue would not get longer by one person.

The same is true, exactly and not merely by analogy, of the machines and lines under a payment system. Ten times the value on the same instructions is the same number of messages to receive, the same number of checks to run, the same number of records to write and the same number of queries to answer later. Every one of those is where the capacity goes, and not one of them counts rupees.

What happens to the average when the value is held still and the count moves?

The control below does one thing: it holds the value carried at Rs 3,60,000 crore and moves only the number of instructions carrying it. Three things happen as it moves: the bar does not change length at any setting, the divisions inside it multiply or thin out, and the marker on the scale beneath slides across two orders of magnitude.

Three readings cover the whole range of the control. At 120 crore instructions, Rs 3,60,000 crore divided by 120 crore instructions gives an average instruction of Rs 3,000.00/-. At 1,200 crore instructions, the default and the year Setu Payments Limited reported, the average is Rs 300.00/-. At 12,000 crore instructions it is Rs 30.00/-. The steps run in factors of ten so that every one of them can be done in the head and the drawing checked rather than trusted, and the reported year sits at the geometric middle of the range rather than at either end.

How does somebody sizing a payment system from the outside actually use this?

Who sets the conditions a payment system operates under?

Four things circled all the way through belong to somebody else. No payment system gets to decide any of them. Each is set by the Reserve Bank of India, each of them moves, and any account that wrote one out would be wrong rather than merely out of date the day it changed.

The second row is the one everything above has been circling: between the instruction and the settlement the money is somewhere, and where it is allowed to be is set rather than chosen. The interval in the netting shape is not just a period of time. The interval holds an obligation that has not been met, and where the money backing that obligation sits, who holds it and on what terms, is exactly what the authority sets. Read the row above the mechanism and the mechanism reads differently.

The other three rows are the conditions under which an arrangement is authorisedPermitted by the authority to operate at all. What has to be true of a party before that permission is given, and what it must keep true afterwards, is set by the authority. to operate at all, what it may charge the banks that are its participantsA bank or other party that is a member of a settlement arrangement and holds an account at its common place. together with every ceiling on that, and what it has to report and how often. None of them is filled in. The sheet, taken to the site named inside it, has its second column filled in from the source. A blank sheet still earns its keep. A condition, and the name of whoever sets it, both outlive whatever number was in force this morning.